FR2944937A1 - Video sequence image's lost pixel blocks restoration method for use during image transmission, involves obtaining restoration motion vector from determined confidence zones, and restoring identified blocks from restoration motion vector - Google Patents

Abstract

The method involves identifying blocks of pixels that are associated with a motion vector from a to- be-restored image of a video sequence. Confidence zones representing selection tolerance of the motion vector associated with the identified blocks are determined for each identified block. A restoration motion vector is obtained from the determined confidence zones. The identified blocks are restored from the restoration motion vector, where the confidence zones are defined by an ellipse. Independent claims are also included for the following: (1) a device for restoring a video sequence (2) an information storage medium for storing computer program instructions for implementing a method for restoring a video sequence (3) a computer program product comprising a set of instructions for implementing a method for restoring a video sequence.

Description

The present invention relates to a method and a device for restoring a video sequence by masking lost areas. The restoration of block-coded video sequences is the action of completing blocks lost during the transmission of a current image. An image of the sequence is divided into square blocks of pixels called macroblocks. A macro-block will be designated in the following by the acronym MB. Block-coded video sequences with a predictive coding such as that defined by the Moving Picture Experts Group (MPEG) consortium, as for example in the MPEG4 - Part 2 and H.264 standards, are considered here. In practice, an MB is usually a square of 16 pixels on the side (with sub-block possibilities, especially for MPEG4 and H.264 formats). In such sequences, the MBs of a current image are calculated by relative translation from the MBs of one or more previously decoded reference images. This process is called a motion compensation method. In the current image, each MB is thus associated with an MB of a reference image by means of a translation vector called motion vector. Since the approximation is imperfect due in particular to the complexity of the real movements, a coding residue is added to the motion compensation to take into account the prediction error between the two MBs. This residue is obtained by subtracting the MB from the current image of the decoded image signal of the associated MB in the reference image and is almost zero if the movement between the reference image and the current image is zero or correctly. approached by a translation.

The set of motion vectors obtained defines, for an image, a motion vector field. The calculation of each motion vector is done at the level of the encoder, this encoder being programmed according to an algorithm allowing the determination of each of these vectors so that it minimizes the associated residue, in order to obtain an optimal compression of the signal. It happens however that during the transmission of the signal certain information is lost, in particular at the level of the motion vectors. Such losses lead to the impossibility of reconstituting some MBs, these MBs then being blocks that are lost. There are already interpolation methods according to which each lost MB is reconstituted by copying an MB of a reference image, pointed by a motion vector obtained by interpolation from motion vectors of other MBs of the image. These methods are known by the generic name of temporal error concealment. These methods are relatively efficient in areas of homogeneous motion, but give inaccurate results in areas of the image where neighboring blocks have non-homogeneous motion vectors.

The aim of the invention is to provide a process for restoring these lost blocks which is both more precise and more reliable. To this end, it proposes a method of restoring a video sequence comprising a plurality of images, at least one of which, called the image to be restored, comprises at least one block of pixels to be restored, as well as several other blocks of pixels. pixels at least some of which are associated with a respective motion vector, each motion vector matching the block of pixels with which it is associated with a block of pixels of a reference image of the video sequence, characterized in that comprises the following steps: identifying, in the image to be restored, blocks of pixels that are associated with a motion vector; determination, for each identified block, of a confidence zone representative of the selection tolerance of the motion vector associated with the identified block; obtaining a restoration movement vector from at least some of these confidence zones thus determined; and restoring said block to be restored from said restoration movement vector. The confidence zones thus determined give information on the selection latitude of each of the motion vectors so as to obtain a restoration motion vector which is as reliable as possible and which minimizes the residue in order to optimize the performance of the compression. Indeed, the determination of zones of confidence around the motion vectors of the identified blocks makes it possible to isolate preferential zones in order to carry out the interpolation of the missing motion vector.

This makes it possible to obtain a restoration motion vector close to the true movement (that is to say the movement that actually occurs in front of the image sensor). This restoration method is also robust and unlikely to be disturbed by the noise of the video sequence which impacts the motion vector calculation. Moreover, the presence of confidence zones makes it possible to exploit only the useful information of the motion vectors which, taken in isolation, sometimes appear incoherent with respect to the true movement because of the fact that the priority is given to the minimization of the costs. compression (i.e. the associated residue) and not the reliability of the vector over the true motion. According to another characteristic, the step of determining a zone of confidence for each identified block comprises a step of obtaining a model representative of the similarity of the identified block with other blocks of 30 pixels, each model being defined by at least one coefficient. The reconstitution (at the decoder level) of a model representative of the similarity during the image restoration thus makes it possible to obtain information on the spatial environment of the pixel blocks in order to determine the corresponding areas of trust. According to other characteristics: the model representative of the similarity of the identified block with other blocks of pixels is based on the calculation of the sum of the absolute value of the differences between said identified block and each said other block; and / or - said other blocks of pixels have at least some pixels in common with said corresponding identified pixel block; Thus, this model representative of the similarity of the identified block with other blocks of pixels characterizes the image to be restored locally around the identified block. According to other features: said at least one coefficient of each model representing the similarity of the identified block with other blocks of pixels is obtained by the least squares method; and / or - each model has the shape of a paraboloid; Advantageously, the model is characterized by a reduced number of coefficients whose calculation is fast. According to other characteristics: each zone of confidence is representative of the shape of the corresponding model; and / or - each zone of confidence is delimited by an ellipse. According to another feature, said step of obtaining a restoration motion vector comprises the steps of: determining, in a parameter space where the motion vectors and the confidence zones are represented, at least one interpolation zone from at least some of these confidence zones; and determining a restoration motion vector from said at least one interpolation zone. Thus, the motion vector obtained is consistent with the set of confidence zones determined for the identified blocks.

According to another characteristic, the step of determining said at least one interpolation zone comprises the step of incrementing each pixel of the parameter space when it is located in a zone of confidence and not retaining as zone (s) interpolation than the zone or zones of this space whose pixel value exceeds a predetermined threshold value. Thus, obtaining the interpolation zone (s) is simple and does not require any additional calculation. According to other features: the step of determining the restoration motion vector comprises, for each interpolation zone, a step of determining an associated interpolation vector; and / or the coordinates of each interpolation vector are obtained by calculating the barycentre of said corresponding interpolation zone. Thus, each interpolation vector is obtained systematically and with a minimum computational cost. According to another characteristic, the step of determining the restoration motion vector comprises, if there are several interpolation zones, a step of combining said corresponding interpolation vectors.

Thus, complex movements can be taken into account. According to another characteristic, among the blocks identified during the identification step, at least some of them are adjacent to the block of pixels to be restored. Thus, the local characteristics of the image to be restored are taken into account. According to other characteristics: the identification step is carried out, for at least some of the blocks, in a search window of predetermined size around the block to be restored; and / or the block identification step includes a step of normalizing the associated motion vectors.

For the same purpose as that indicated above, the present invention also relates to a device for restoring a video sequence comprising a plurality of images of which at least one of them, called the image to be restored, comprises at least one block. pixels to be restored as well as several other blocks of pixels at least some of which are associated with a respective motion vector, each motion vector matching the block of pixels with which it is associated with a block of pixels of a reference image of the video sequence, characterized in that it comprises: means for identifying, in the image to be restored, blocks which are associated with a motion vector; means for determining, for each block identified, a confidence zone representative of the selection tolerance of the motion vector associated with the identified block; means for obtaining a restoration motion vector from at least some of these confidence zones thus determined; and means for restoring said block to be restored from said restoration movement vector. Still for the same purpose, the present invention also aims at a means for storing information, readable by a computer or a microprocessor retaining instructions of a computer program, characterized in that it allows the implementation of a restoration process as explained above. Still for the same purpose, the present invention also aims at a computer program product, which can be loaded into a programmable device, characterized in that it comprises sequences of instructions for implementing a restoration method such as discussed above, when this program is loaded and executed by the programmable apparatus. Since the particular features and advantages of the recovery device, the information storage means and the computer program product are similar to those of the restoration method, they are not recalled here.

The description of the invention will now be continued by the detailed description of an exemplary embodiment, given below by way of illustration but without limitation, with reference to the appended drawings, in which: FIG. 1 is a flowchart illustrating the main steps of a method of restoring a video sequence according to the present invention, in a particular embodiment; - Figure 2 schematically shows the identification of blocks around the block of pixels to be restored; FIG. 3 illustrates for three blocks of pixels of an image the values of similarity matrices from which confidence zones are obtained for each of these blocks; FIG. 4 shows a schematic view illustrating a parameter space in which the motion vectors of the blocks identified to restore a lost block are represented as well as their respective confidence zones; and FIG. 5 is a schematic representation of a particular embodiment of a device capable of implementing the present invention. In the present description, the video sequences are coded in P-mode blocks with a predictive coding such as that defined by the Moving Picture Experts Group (MPEG) consortium, as for example in the MPEG4 - Part 2 and H.264 standards. Each MB is here a square of 16 pixels side (with possibilities of sub-blocks especially for MPEG 4 and H.264). Prior to the presentation of the restoration method according to the invention, reference will here be made in more detail to the operating principle of the image acquisition method by motion compensation at the coding of the image sequence. As previously discussed, such a method involves associating an MB of the current image with an MB of an earlier image via a motion vector and its residue.

It is important to try to minimize the residual for each block encoded by motion compensation, as this determines the quality of the compression that will be performed at the encoder. Indeed, the lower the energy of the residue signal (the energy can be calculated as equal to its variance for example), the more it can be compressed. An example of an algorithm for optimally determining the motion vector minimizing the associated residue is based on the study of the similarity of the block of the current image to be coded with blocks of the reference images, each block of images reference being associated with a candidate motion vector. The encoder thus seeks a motion vector among several candidate motion vectors such that the pointed block of the reference image is as close as possible to the block to be encoded of the current image. It should be noted that motion vectors can be calculated with sub-pixel precision: precision is / 2 pixel for MPEG4 - Part 2 and 1/4 pixel for H.264. A non-integer motion vector corresponds to an interpolated reference MB of a pixel fraction, according to the standard used, in order to make possible similarity calculations with the MB of the current image. The advantage of calculating motion vectors with sub-pixel precision is to reduce the amplitude of the residues. The most commonly used method of estimating the similarity between two blocks is that based on the sums of the absolute values of the differences between the block considered and the other blocks (SAD: sum of absolute difference).

The SAD is zero if the two blocks are identical and it is important if these two blocks differ a lot. The encoder thus searches for the SAD which has the lowest value and selects the corresponding motion vector so that the blocks thus associated are as similar as possible to one another. The encoder is therefore responsible for calculating the SAD of the different candidate blocks and searching for an absolute minimum. The set of SAD values associated with the blocks define a surface called the similarity map or the SAD map. Around a minimum, this map can be modeled by a paraboloid whose shape depends on the nature of the pixels of the blocks to be compared.

The local shape of the similarity map or "SAD error shape" around the minimum detected by the encoder thus gives an indication of the nature of the MBs. For example, if the MBs being compared are very textured, then the shape of the similarity map around the minimum is very narrow (the similarity map values around the minimum are much higher than this minimum). On the other hand, for MBs with almost constant pixels, the shape of the similarity map around the minimum is very wide (the values of the similarity map around the minimum are very slightly higher than this minimum).

For blocks overlapping a contour of the same direction, the shape of this map is stretched in that direction. The position of the minimum of the SAD map thus indicates the optimal motion vector of the block whereas the shape of the SAD map gives an indication of the nature of the pixels of the block to be encoded.

The restoration method according to the invention will now be described using the algorithm of FIG. 1 which illustrates the main steps 11 to 15 of this method. Such an algorithm can be executed in the device of FIG. 5. Prior to the actual restoration process, the decoder informs in a prior step 10 of the loss of the MBs. Each MB being in the illustrated example consisting of a square of 16x16 pixels, this information is saved in a matrix 16 times smaller than the coded images, each cell of this matrix indicating whether the corresponding MB is lost or valid. In the example illustrated in FIG. 2, the three hatched blocks are detected as being lost in step 10. For each lost block thus detected, the restoration method, the implementation of which begins in the following step 11, identifying in the image to be restored, among the blocks which are associated with a motion vector (that is to say among the blocks that are valid), blocks of pixels neighboring the lost pixel block considered. By way of example, in FIG. 2, for the lost MB 200, the received blocks 202, 204 and 206 are identified.

The purpose of this identification step is to use motion vectors of MBs at least some of which are close to the lost MB so that the interpolation of the motion vector for the lost block is the most reliable possible. For this purpose, for example, for the search window, all the received MBs which are distant from an MB lost by at most two MBs are retained.

Each motion vector of each identified neighbor MB corresponds to a reference image which may vary from one MB to another depending on the video format used. Here, therefore, it is also decided to standardize in this step the motion vectors so that they only point to a single image preceding the current image. For this, the coordinates of a motion vector are divided by the number of images that are between the current image and the reference image of this vector. This normalization makes it possible to interpolate motion vectors that do not have the same reference images. In the next step 12, for each identified block, a confidence zone associated with the corresponding motion vector is determined. This confidence zone is directly deduced as will be seen below from a similarity model (similarity map) of the corresponding motion vector and is an indicator of the nature of the associated pixel block.

Before giving more details on this zone of confidence, we will first describe how it is obtained from the corresponding similarity map. At the level of the decoder where the restoration method is implemented, the similarity map of each motion vector is not known. It is therefore necessary first of all to recalculate it before being able to deduce the corresponding zone of confidence.

This similarity map can be calculated in many ways, the method chosen here is based on the SAD estimator.

In order to speed up the calculations, the SAD map is calculated on the identified block according to itself and its immediate neighborhood (this is called a self-similarity map because the candidate blocks selected for the calculation have a large number of pixels in common with the main block).

In the illustrated example, this map is thus obtained by comparing a part of the contents of this MB with itself (center of the matrix), as well as with the MBs which are shifted respectively by +/- 1 pixel according to horizontally and / or offset by +/- 1 pixel depending on the vertical. In the case of a directly adjacent (or adjacent) MB of a lost MB, a subset of pixels of this MB is used for the calculation of self-similarity, to take into account the fact that some pixels on one of the edges of the MB are missing.

Consider the macroblock or the sub-macroblock M of size nxm pixels. We thus define the map of self-similarity of M by a matrix S (i, j) based on the estimator of SAD in the following way: n-1 m-1 S (i, j) = LLM (x, y ) -M (x + 1 + i, y + 1 + j) x = 1 y = 1 with i and j = - 1, 0 or 1 (the matrix S being a 3x3 matrix).

The center S (0, 0) of the matrix is equal to 0 since for i = 0 and j = 0 the MB is compared to itself.

The other eight values are greater than or equal to 0.

This self-similarity calculation makes it possible to simplify the calculations while ensuring that the similarity map thus obtained can be modeled around its extremum, which ensures a high reliability in the determination of the

25 corresponding confidence zone.

Figure 3 illustrates different self-similarity maps according to different areas of an image.

For each map, the nine values of S (i, j) are modelized by a paraboloid:

The six parameters A, B, C, D, E, F are computed by the least squares method knowing the nine elements. values of the matrix S (i, j). From these six parameters, are extracted the three parameters of the zone of confidence (8, a and b) according to the following system of equation: 1 (C 0 = ùArc tan 2 AùBi 1 a = (A.cosO + B.sin0 çC.cos0.sin0) 2 b = (A.sinO + B.cosO + C.cosO.sin 0) 2 These confidence zones are thus delimited by the ellipse (characterized by three parameters, namely: the orientation 8 , the half major axis has 10 and the half minor axis b) defined above and whose center is coincident with the end of the associated motion vector.The shape of the ellipse delimiting this zone is therefore directly correlated to the shape of the corresponding similarity map, the three parameters of the zone of confidence reflecting the nature of the pixels included in the block M (homogeneous, textured and / or elongated) .In this zone of confidence the motion vector can be replaced by any another vector of motion with a near-zero influence on the amount of information of the residue and therefore on the rate of comp Thus, for a MB included in a textured area the paraboloid is very sharp and the parameters A and B are large, which has the consequence that the zone of confidence has small half-axes (the zone of confidence). has a weak surface). This vector is therefore close to the true motion, that is to say the movement that actually occurs in front of the image sensor. Such a motion vector must therefore be considered differently from a motion vector associated with an MB of a homogeneous zone. In fact, by contrast, for an MB of a homogeneous zone the parameters A and B are small and the zone of confidence has important half-axes 1. In such a case, the associated SAD card is flat and its zone of confidence is very wide. Such a motion vector does not need to be chosen very precisely and can be exchanged for another motion vector with almost no influence on the amount of information of the residue and thus on the compression ratio of the MB. An intermediate case concerns the MBs overlapping a strong contour, the SAD card being in this case elongated in the direction of the contour. In other words, the motion vector can be replaced by any other motion vector such that the pointed MB overlaps the contour in the same way. For such MBs, the motion vectors are generally very different from the true motion. Nevertheless, the component of the motion vector in the direction perpendicular to that of the contour is very precise and similar to the true motion. For such MBs, the zone of confidence is therefore an ellipse stretched in the direction of the contour and very flattened in the direction perpendicular to the contour. In other words, a zone of confidence is representative of the selection tolerance of the motion vector associated with the block considered. This information is used in the present invention to provide an indication of the room for maneuver in determining the component of the motion vector to be interpolated that depends on that particular motion vector. Indeed, if the zone of confidence for this vector is large, the margin of maneuver in the determination of the component of the motion vector to be interpolated which depends on this motion vector will be important whereas it will be narrower if the zone of confidence is reduced. In the next step 13, the confidence zones of each identified block are then accumulated in a parameter space in which each of the corresponding motion vectors is also represented, in order to isolate one or more preferential zones for the choice of the vector to be interpolated. .

The parameter space is an image in which all the pixels are initially set to zero and in which the motion vectors are represented. The size of the parameter space is defined by the desired amplitude and resolution. Let M be the size of the parameter space and s the resolution factor, a motion vector is converted into coordinates (M / 2 + x.s, M / 2 + y. $) In the parameter space. The resolution factor s makes it possible to define the sampling step of the motion vectors.

The same scaling is performed for the coefficients a and b of each ellipse according to a specific resolution factor s'. The confidence zones are thus accumulated in the parameter space. FIG. 4 shows three motion vectors 21 to 23, each associated with a neighboring block of the same block to be restored. Each motion vector corresponds to an ellipse 24 to 26 delimiting the associated confidence zone. Each motion vector points to the center of the ellipse with which it is associated. Each pixel of this space is incremented by the value 1 as soon as it is located in a zone of confidence, that is to say within an ellipse defined as before and projected in this parameter space. Since all the confidence zones are thus represented in the parameter space, the accumulation of the confidence zones reveals one or more areas in recovery. These overlapping areas correspond to areas where the pixels reach maximum values (also referred to as accumulation peaks) and whose position indicates the potential coordinates of an interpolated motion vector. Indeed, it is in these areas of overlap that the chances are optimal to obtain interpolated motion vectors, each of which corresponds reliably to at least a part of the true motion. In the illustrated example, three zones of confidences are represented, but it will be noted that it is possible in other cases to have less than three confidence zones in the parameter space, or even more than three confidence zones. the principle of the invention remaining identical.

The next step 14 (calculating a recovery vector for each lost block) consists in isolating or extracting these overlapping zones in order to determine interpolated vectors from these zones. For this reason these areas will be called later interpolation zones.

In the example shown, the method used to isolate the overlapping areas from the parameter space is as follows:

- detection of the maximum value pixel in the parameter space;

thresholding of the parameter space with respect to the threshold equal to max. The pixels of lower value are set to 0 (in the example illustrated in FIG. 4, this thresholding makes it possible to isolate as an interpolation zone the zone which is hatched, situated at the intersection of the three ellipses); - Insulation interpolation areas, these areas being defined by a set of related pixels greater than the threshold determined above. If several zones are observable it means that the true movement is made of several combinations of movements;

for each interpolation zone, a barycentre is calculated to determine the coordinates of the corresponding interpolated vector.

The center of gravity is calculated for each zone Z; (i varying from 1 to n where n corresponds to the total number of interpolation zones obtained) by the order moment: L xP (x, y) L yP (x, y) - (x, Y) EZi - ( x, yEZt x` LP (x, y) 'y ~ - LP (x, y) (x, yE Z, (x, yE Z, with P (x, y) the value of the pixel of the space of parameters at the position (x, y) For each zone Z ;, the coordinates (V (i) X, V (i) y) of the vector of

corresponding interpolated motion is expressed as a function of the size and the resolution factor of the parameter space: (x ~ ûM) (Yi-M) V (i) x = 2; V (i) y = 2 The number of motion vectors thus calculated is a function of the motion occurring in the video sequence. Thus, if the movement is a homogeneous translation then a single interpolation zone appears, this zone corresponding to a single interpolated motion vector. On the other hand, if the movement is multiple (for example in the presence of several sources having different movements in the image), then several interpolation zones are observed corresponding to as many interpolated motion vectors.

In the case where only one interpolation zone appears, the final motion vector, also called restoration motion vector, corresponds to the single interpolated motion vector.

In the case where several interpolation zones appear, it is necessary to provide here an additional step which consists in determining the restoration motion vector from the set of interpolated motion vectors at each of the interpolation zones.

This vector is determined by combining the interpolated vectors as follows:

from the identified MBs, those which are the closest to the MB to be restored are selected;

for each identified MB thus selected, the interpolated vector that is closest to its motion vector is searched for; and

the restoring motion vector is then determined by averaging the interpolated motion vectors as well as

25 selected.

Finally, in a last step 15 to replace the lost block, a simple motion compensation is applied from the restoration motion vector. It is simply a matter of copying all the pixels of the block of the reference image which is pointed by the vector of

30 restoration movement on the lost block of the current image, the lost MB being thus restored.

It should be noted that the reference image is not necessarily an earlier image of the video sequence. In the case where the restored image comprises several lost blocks, the method according to the invention can be applied to all the blocks lost simultaneously or even in several times on subsets of the lost blocks such that all these subsets cover all blocks lost once. FIG. 5 shows a particular embodiment of an information processing device able to function as a device for restoring a video sequence according to the present invention. The device illustrated in FIG. 5 may comprise all or part of the means for implementing a restoration method according to the present invention, and which correspond in particular to means making it possible to implement each step of FIG. namely: means for detecting in the image the lost block or blocks; means for identifying in this image, for each lost block, received blocks which are associated with a motion vector; means for determining for each identified block its zone of confidence; Means for obtaining a restoration motion vector from at least some of these confidence zones thus determined; and means for restoring each lost block from the corresponding restoration motion vector. According to the embodiment chosen, this device can for example be a microcomputer or a work station 100 connected to different peripherals, for example a digital camera 101 (or a scanner, or any other means of acquisition or image storage) connected to a graphics card (not shown) and thereby providing information to be processed according to the invention. The microcomputer 100 preferably comprises a communication interface 102 connected to a network 103 capable of transmitting digital information. The microcomputer 100 also includes a permanent storage means 104, such as a hard disk, as well as a reader of temporary information storage means such as a floppy disk drive 105 for cooperating with a floppy disk 106.

The diskette 106 and the hard disk 104 may contain software layout data of the invention as well as the code of the computer program (s) whose execution by the microcomputer 100 implements the present invention. , this code being for example stored on the hard disk 104 once it has been read by the microcomputer 100.

In a variant, the program (s) (based on the algorithm of FIG. 1) enabling the device 100 to implement the invention are stored in a read-only memory (for example of the ROM type) 107. According to another alternatively, this or these program (s) are received totally or partially through the communication network 103 to be stored as indicated. The microcomputer 100 also comprises a screen 109 for displaying the information to be processed (for example the parameter space shown in FIG. 4) and / or serving as an interface with the user, so that the user can, for example, to set certain processing modes using the keyboard 110 or any other appropriate means of pointing and / or input such as a mouse, an optical pen, etc. A calculation unit or central processing unit (CPU) 111 executes the instructions relating to the implementation of the invention, these instructions being stored in the ROM 107 or in the other storage elements described. In particular, the central processing unit 111 is adapted to implement the algorithm illustrated in FIG. 1. When the device 100 is powered up, the programs and processing methods stored in one of the non-volatile memories, for example the ROM 107, are transferred into a random access memory (for example of the RAM type) 112, which then contains the executable code of the invention as well as the variables necessary for the implementation of the invention.

Alternatively, the method of restoring the digital signal can be stored in different storage locations. In general, a means for storing information that can be read by a computer or by a microprocessor, whether or not integrated into the device, possibly removable, can store one or more programs whose execution implements the method of restoring the data. a video sequence described above. The particular embodiment chosen for the invention can be modified, for example by adding updated or improved processing methods; in such a case, these new methods can be transmitted to the device 100 by the communication network 103, or loaded into the device 100 via one or more floppies 106. Of course, the floppies 106 can be replaced by any information medium deemed appropriate (CD-ROM, memory card, etc.). A communication bus 113 allows communication between the different elements of the microcomputer 100 and the elements connected thereto. Note that the representation of the bus 113 is not limiting. Indeed, the CPU 111 is, for example, capable of communicating instructions to any element of the microcomputer 100, directly or through another element of the microcomputer 100.

According to a first variant, each SAD card is not determined on the MBs received from the image to be restored but on the corresponding MBs indicated by the associated motion vectors of the image or corresponding reference images. According to a second variant, each similarity map is obtained by other methods than by SAD calculations. According to a third variant, the blocks identified to carry out the restoration of the lost block are not necessarily close to the lost block, for the purpose for example of having more motion vectors and therefore more confidence zones in order to better highlight the 30 accumulation peaks. According to a fourth variant, the zones of confidence are obtained from the modelings of the similarity maps, not in the form of elliptical surfaces but under any other type of surface shape or even in the form of segments (segments indicating in this case only a preferential direction for the interpolation of the restoration motion vector).

Many other variants are possible depending on the circumstances, and it is recalled in this regard that the invention is not limited to the examples described and shown.

Claims (21)

REVENDICATIONS1. A method of restoring a video sequence comprising a plurality of images, at least one of which, called the image to be restored, comprises at least one block of pixels to be restored, as well as several other blocks of pixels, at least some of which are associated with a respective motion vector, each motion vector matching the block of pixels with which it is associated with a block of pixels of a reference image of the video sequence, characterized in that it comprises the following steps: identifying (11), in the image to be restored, blocks of pixels that are associated with a motion vector; determination (12), for each identified block, of a confidence zone representative of the selection tolerance of the motion vector associated with the identified block; obtaining (13, 14) a restoration movement vector from at least some of these confidence zones thus determined; and restoring (15) said block to be restored from said restoration movement vector. 2. Method according to claim 1, characterized in that the step of determining a zone of confidence for each identified block comprises a step of obtaining a model representative of the similarity of the identified block with other blocks of pixels, each model being defined by at least one coefficient. 3. Method according to claim 2, characterized in that the model representative of the similarity of the identified block with other blocks of pixels is based on the calculation of the sum of the absolute value of the differences between said identified block and each said other block. 4. Method according to any one of claims 2 or 3, characterized in that said other blocks of pixels have at least some pixels in common with said corresponding identified pixel block. 5. Method according to any one of claims 2 to 4, characterized in that said at least one coefficient of each model representative of the similarity of the identified block with other blocks of pixels is obtained by the least squares method. 6. Method according to any one of claims 2 to 5 characterized in that each model has the shape of a paraboloid. 7. Method according to any one of claims 2 to 6, characterized in that each confidence zone is representative of the shape of the corresponding model. 8. Method according to any one of claims 1 to 7, characterized in that each zone of confidence is delimited by an ellipse. 9. Method according to any one of claims 1 to 8, characterized in that said step of obtaining (13, 14) a restoration motion vector comprises the steps of: - determination (13), in a space parameters representing the motion vectors and the confidence zones of at least one interpolation zone originating from at least some of these confidence zones; and determining (14) a restoration motion vector from said at least one interpolation zone. Method according to claim 9, characterized in that the step of determining (13) said at least one interpolation zone comprises the step of incrementing each pixel of the parameter space when it is located in a zone of confidence and to retain as zone (s) of interpolation only the zone or zones of this space whose pixel value exceeds a predetermined threshold value. 11. Method according to any one of claims 9 or 10, characterized in that the step of determining (14) the restoration motion vector comprises, for each interpolation zone, a step of determining a vector d associated interpolation. 12. Method according to claim 11, characterized in that the coordinates of each interpolation vector are obtained by calculating the barycentre of said corresponding interpolation zone. 13. Method according to any one of claims 11 or 12, characterized in that the step of determining (14) the restoration motion vector comprises, if there are several interpolation zones, a step of combining said vectors. corresponding interpolation. 14. Method according to any one of claims 1 to 13, characterized in that, among the blocks identified during the identification step, at least some of them are adjacent to the block of pixels to be restored. 15. The method as claimed in claim 14, characterized in that the identification step is performed, for at least some of the blocks, in a search window of predetermined size around the block to be restored. 16. Method according to any one of claims 1 to 15, characterized in that the step of identifying blocks comprises a step of normalizing the associated motion vectors. 17. Device for restoring a video sequence comprising a plurality of images of which at least one of them, called the image to be restored, comprises at least one block of pixels to be restored as well as several other blocks of pixels, at least some are associated with a respective motion vector, each motion vector matching the block of pixels with which it is associated with a block of pixels of a reference image of the video sequence, characterized in that it comprises: means for identifying, in the image to be restored, blocks that are associated with a motion vector; means for determining, for each block identified, a confidence zone representative of the selection tolerance of the motion vector associated with the identified block; means for obtaining a restoration movement vector from at least some of these confidence zones thus determined; and means for restoring said block to be restored from said restoration movement vector. 18. Device according to claim 17, characterized in that said determination means for each identified block of a confidence zone are adapted to obtain a model representative of the similarity of the identified block with other blocks of pixels. 19. Device according to any one of claims 17 or 18, characterized in that said determination means for each identified block of a confidence zone are adapted to obtain said model representative of the similarity of the identified block with other blocks pixels by calculating the sum of the absolute value of the differences between said identified block and each said other block. 20. A computer-readable information storage medium or a microprocessor holding instructions of a computer program, characterized in that it allows the implementation of a restoration method according to any one of claims 1 to 16. 21. Computer program product that can be loaded into a programmable device, characterized in that it comprises sequences of instructions for implementing a restoration method according to any one of claims 1 to 16, when this program is loaded and executed by the programmable device.